What should I change for this athlete this week?
The job is not to stare at another score. The job is to find the likely lever, check whether the evidence is strong enough, and turn it into the next training call.
A coach rarely needs more data in the moment. They need a cleaner answer to a narrower question: what should I change for this athlete this week?
That question is harder than it sounds. Sleep, load, HRV, soreness, travel, missed sessions, nutrition, and benchmark history all compete for attention. A dashboard can show all of it and still leave the coach doing the hard part by feel.
AthDash is built around the missing middle: turning athlete exports into a defensible coaching decision, with the uncertainty still attached.
01Start with the decision, not the dashboard
The weekly coaching call is usually practical. Do we add intensity, pull volume, hold the plan, move the benchmark, or ask a different recovery question? The data layer should work backward from that decision.
That means the output cannot just be "readiness is 73" or "sleep is down." It has to say what relationship appears to matter for this athlete, what outcome it is tied to, and whether the evidence is strong enough to coach from.
- Question
- What should change before this week's benchmark session?
- Likely lever
- Keep aerobic decoupling below the athlete's own late-session threshold.
- Outcome
- Next time-trial performance, adjusted for fatigue and recent load.
- Evidence
- Supported relationship, useful sample, interval clears the no-effect line.
- Caveat
- Do not generalize to strength work; this finding is about endurance benchmark sessions.
- Action
- Hold intensity, trim the final endurance block if decoupling rises, and retest after a cleaner aerobic finish.
The coach still makes the call. AthDash just makes the evidence easier to inspect before the call gets made.
02Separate candidates from claims
Most athlete data starts as a list of candidates. Sleep might matter. Acute load might matter. HRV might matter. Soreness might matter. A previous benchmark might matter. Listing candidates is useful, but it is not a claim.
A claim needs a higher bar. It should name the driver and outcome, show the effect estimate, show the interval, show how much athlete-specific data supports it, and say what would make the claim too thin to use.
That is where a lot of coaching tools blur the line. They turn every signal into an insight. AthDash keeps the gate visible, including when the right answer is not enough evidence yet.
03Ask what likely moved the outcome for this athlete
The phrase "for this athlete" does real work. A population average can be helpful background, but the training adjustment has to land on one body with one history. A signal that predicts one athlete's benchmark may be noise for another.
So the useful answer is not "sleep matters for athletes." It is closer to: when this athlete's sleep regularity improved before benchmark sessions, their time-trial output tended to improve, after accounting for recent load and fatigue. If that relationship is thin, the output should say so.
This is also why effect modifiers matter. The same load can help under one condition and hurt under another. The coach needs the condition, not just the average.
04Turn uncertainty into usable wording
A coach does not need a statistics lecture during a roster review. They do need to know how strongly they can speak. AthDash treats the evidence state as part of the result, not a footnote.
- Supported means the relationship is strong enough to use in the decision, with caveats included.
- Exploratory means the pattern is worth testing, not presenting as settled.
- Borrowed means a cohort pattern can guide observation, but it is not yet this athlete's own effect.
- Insufficient means the data is too thin, even if the chart looks tempting.
AthDash does not just point to a number. It tells the coach how much permission that number has earned.
05Let the agent read the same packet before it writes
This is where the same output becomes useful for AI coaching agents. The agent does not need to re-read every workout row, re-build every benchmark, or guess which caveat belongs in the athlete reply. It can ask AthDash for the compact packet first.
That saves context and token budget, but the bigger value is accuracy. The language model can spend its attention on tone, timing, and the athlete conversation while deterministic code handles the evidence gate.
The result is a cleaner division of labor: code checks what the data supports, the coach decides what to do, and the agent repeats only the licensed version back to the athlete.
06The weekly answer should be short
A good decision packet should fit inside the coaching workflow. It should be short enough to read during roster review and specific enough to defend later.
The best version sounds almost boring: here is the likely lever, here is why we think it matters, here is how strong the evidence is, here is the caveat, and here is the training adjustment that follows. That is the point. The output is useful because it reduces the decision to the part a coach can act on.